PBE

PBE encodes protein three-dimensional structures into a 16-state Protein Blocks (PBs) structural alphabet and performs PB-based prediction, comparison, and database mining to analyze local folding patterns defined by the φ and ψ dihedral angles of five consecutive C(alpha) atoms.

Key Features:

• Structural Alphabet (Protein Blocks): Sixteen distinct PBs represent local folding patterns defined by φ and ψ dihedral angles of five consecutive C(alpha) atoms, derived from unsupervised cluster analysis of PDB structures.
• Successive PB Dependence: The dependence between successive PBs is explicitly considered to improve local structural characterization.
• Bayesian Amino-Acid Propensity Prediction: A Bayesian approach predicts amino-acid propensities relative to PBs with ~35% success, rising to over 75% when sequence windows are grouped into "sequence families".
• Prediction Strategies: Two strategies permit (1) determining the number of PBs required per site for a target accuracy and (2) identifying sites predictable with a fixed number of blocks at a chosen accuracy.
• Encoding and Alignment: Protein 3D structures are encoded into PB sequences and aligned using dynamic programming with a PB-specific substitution matrix.
• Tool Components: PBE-T transforms PDB files into PB sequences; PBE-ALIGNc compares two protein structures via PB alignments; PBE-ALIGNm mines the SCOP database for similar structures.
• Database Integration: PBE-SAdb contains preprocessed PB sequences from SCOP at 95% identity and all-against-all pairwise PB alignments across family and superfamily levels.

Scientific Applications:

• Ab initio Protein Modeling: PB-based sequence–structure dependencies inform and improve ab initio protein modeling and local conformation sampling.
• Structure Comparison: PB sequence encoding enables comparative analysis of protein conformations and identification of structurally similar regions.
• Database Mining: Mining SCOP via PB alignments facilitates discovery of related structures at family and superfamily levels.
• Structure Prediction Refinement: PB-based propensity predictions and sequence-family grouping enhance the accuracy of local structure predictions.

Methodology:

Unsupervised cluster analysis of PDB structures to derive 16 PBs; explicit modeling of dependence between successive PBs; Bayesian estimation of amino-acid propensities and grouping of sequence windows into sequence families; encoding of 3D structures into PB sequences and dynamic programming alignment using a PB-specific substitution matrix; transformation of PDB files via PBE-T and storage of preprocessed PB sequences and all-against-all PB alignments in PBE-SAdb.